Notes:

Change in frequency, higher frequency at bifurfcation.

Need to remake ramp figures or get permission to reprint
Need to remake spec figures, or get permission

Evidence of partial propagation. Amplitude attenuation as a function of distance from electrode during passive stim state (we have it maybe for the
movement state). 

Single example, cherry picked! 

1: it is a hopf (because data)
2: it can't be the other 2 SNIC, Saddle-node
3: how do we get excitatbility in a hopf?
4: this is how: put the system close to bifurcation and make nullclines excitable / close together (what parameter regime is needed)
5: problem: waves propagate indefinitely
-- Is this consistent with the data? 
-- -- Re-entrant waves never observed after stim removal
-- -- Immedediate end of gamma power within one cycle
-- -- Amplitude attenuation further from stim site
6: Infinite propagation seemingly inconsistnat with data?
-- Makes sense, because cortex had mechanism to avoid spurious high-rate / wave activity (don't want a seizure)
7: Can we fix the model?
-- yes, and here is the parameter regime
-- Here is how you prove the region of paramemeter that give finite propagation
8: How realistic is it?
-- It seems pretty fine-tuned, what are the implications?
-- Does the brain self tune?
-- Model is VERY simplified, in reality MANY E and I cell types
-- Discuss: could there be fast-timescale network homeostatic mechanism:
-- -- inbuilt circuity to  maintain function even though
-- -- -- rates are perturbed
-- -- -- oscillations emerge
-- Open question: simple reduce E-I model is fine tuned, 
-- -- Are more complex ("accurate") models easier to tune? 
9: Predictions of model for future experiments.
-- Fine tuning? (can we collapse parameter space down to 1 or 2 bifurcation parameters and dicuss how these are tuned?)

Then: discuss how this related robustness and homeostasis